paper-writing（论文写作）

Paper Writing

A systematic 11-step workflow for writing academic papers, with section-specific templates and battle-tested writing principles.

When to Use This Skill

• User asks to write or draft a paper or paper section
• User needs LaTeX templates for Abstract, Introduction, Method, Experiments, etc.
• User wants to improve academic writing quality
• User mentions "paper writing", "write introduction", "draft method section", etc.

Artifact Sources

If you used upstream EvoSkills, pull these artifacts before writing:

The 11-Step Writing Process

Follow these steps in order. Each step builds on the previous one.

1. Draw a pipeline figure sketch — Sketch the method's pipeline figure to clarify the overall approach. The figure highlights novelty, not just explanation.
2. Design the story and plan experiments — Outline the paper's story (core contribution, module motivations). List comparison experiments and ablation studies. Draft an Introduction writing plan.
3. Write Method — Organize the Method writing plan, then draft Method. Run experiments in parallel.
4. Revise Introduction and Method — Iterate on both sections while experiments continue.
5. Write Experiments — Once experiments are mostly done, organize the Experiments writing plan, then draft.
6. Polish figures — Finalize the pipeline figure. Create the teaser figure.
7. Write Related Work — List related papers, group into topics, write paragraphs.
8. Review the paper — Self-review Introduction, Method, and Experiments. Use the paper-review skill.
9. Write Abstract — Organize the Abstract writing plan, then draft.
10. Choose the title — List important keywords, then compose an informative title.
11. Iterate — Repeatedly review and revise the entire paper.

Counterintuitive Writing Rules

Apply these rules when aiming for higher acceptance probability:

1. Underclaim in prose, overdeliver in evidence: Reduce adjective intensity in Abstract/Introduction; let tables and figures carry the strength.
2. State one meaningful limitation early: A controlled limitation statement increases credibility and lowers reviewer suspicion.
3. Lead with mechanism, not only metric: Explain why the method works before listing numbers; reviewers trust causal logic more than isolated gains.
4. Prefer one decisive figure over many average figures: Build one "cannot-ignore" figure that validates the central claim under hard conditions.
5. Remove weak but flashy claims: Any claim without direct evidence should be deleted, even if it sounds impressive.
6. Declare scope boundaries explicitly: One sentence in Introduction and Conclusion stating what your method targets reduces reviewer fear of hidden assumptions.
7. Show one failure case: Include one representative failure with diagnosis — it signals competence, not weakness.

See references/counterintuitive-writing.md for all 7 tactics with before/after examples.

Section Quick Reference

Abstract

Answer these questions before drafting:

1. What technical problem do we solve, and why is there no well-established solution?
2. What is our technical contribution?
3. Why does our method fundamentally work?
4. What is our technical advantage / new insight?

Three template versions: challenge-first, insight-bridge, multi-contribution. See references/abstract-templates.md

Introduction

Thinking process (reverse then forward):

• Reverse: (1) What is the technical problem? (2) What are our contributions? (3) Benefits and new insights? (4) How to lead into the challenge?
• Forward: (1) Task → (2) Previous methods → challenge → (3) Our contributions → (4) Technical advantages and insights

Four ways to introduce the task, three ways to present challenges, four ways to describe the pipeline. See references/introduction-templates.md

Anti-pattern: Never write "here is a naive solution, then our improvement" — this makes the work appear incremental.

Method

Every pipeline module needs three elements:

1. Module design — Data structure, network design, forward process (given X input, step 1..., step 2..., output Y)
2. Motivation — Why this module exists (problem-driven: "A remaining challenge is...")
3. Technical advantages — Why this module works well

Start with an Overview paragraph (setting + core contribution + section roadmap), then one subsection per module. See references/method-templates.md

Experiments

Three key questions to answer:

1. How to prove our method is better → comparison experiments
2. How to prove our modules are effective → ablation studies
3. How to showcase the method's upper limit → demos on challenging data

Ablation studies need: one big table (core contributions) + several small tables (design choices, hyperparameters). See references/experiments-guide.md

Related Work

Three-step process:

1. List papers closely related to our method (most important — missing key references can cause rejection)
2. Determine topics based on research direction and algorithm techniques
3. Organize writing plan based on listed papers

See references/related-work-guide.md

Conclusion

• Must include Limitation section (reviewers frequently cite "no limitation" as a weakness)
• Limitation = task goal / setting limitations (like future work), NOT technical defects
• Rule: "If our method does not fall below current SOTA metrics, it is not a technical defect"

Supplementary Material

For page-limited venues, decide what goes in main paper vs. supplementary:

• Core evidence for claims must stay in the main paper
• Implementation details, extra ablations, full visual galleries go in supplementary
• Reference supplementary at the point of need, not as a blanket statement

See references/supplementary-guide.md

Core Writing Principles

1. One message per paragraph — Each paragraph conveys exactly one point
2. Topic sentence first — The first sentence tells readers what this paragraph is about
3. Plan before writing — Outline the writing plan, refine each part, then write English sentences
4. Flow between sentences — Ensure logical continuity between consecutive sentences
5. Terminology consistency — Use the same term throughout; do not alternate names
6. Reverse-outlining — After writing, extract the outline from paragraphs; check if the flow is smooth
7. Iterate relentlessly — Polish repeatedly, asking whether readers can follow

See references/writing-principles.md

Key Insight

Visual polish directly influences review outcomes. See the paper-planning skill's figure-design.md for the full visual quality guide.

Paper Title Guidelines

• The title attracts specific reviewers — choose keywords carefully
• Before writing the title, list important keywords, then compose
• Title must be informative: include the technique, task, or problem solved
• Avoid generic titles; specific phrases are more memorable

LaTeX Assets

• assets/paper-skeleton.tex — Annotated LaTeX skeleton with section structure
• assets/table-style.tex — Booktabs table macros with color highlighting

Handoff to Review

Before invoking paper-review, verify this checklist:

• All sections (Abstract, Introduction, Method, Experiments, Related Work, Conclusion) drafted
• Every claim in Abstract/Introduction anchored to a table or figure
• Limitation section present in Conclusion
• Pipeline figure and teaser figure finalized
• All \todo{} markers resolved or removed

Section Navigation